What counts as cheating when your child uses AI for homework

Your child hands over a paragraph that is better than anything they wrote last term. Tidier. Oddly confident. You have about four seconds to decide what kind of parent you are going to be about it, and both obvious moves are bad. Accuse them and you might be wrong, and you have taught them that good work invites suspicion. Say nothing and you have quietly agreed that this is fine.

Britain has been working on that exact problem at national scale, and in March it published something more useful than a ban. A procedure.

What the exam boards actually ask for

On 9 March 2026, Ofqual, the exams regulator for England, put out a set of classroom resources called “AI in coursework”: a lesson plan for teachers and a briefing pack for school leaders. They sit on top of the real rulebook, which belongs to the Joint Council for Qualifications, the body speaking for eight exam boards including AQA, OCR, Pearson, WJEC and CCEA. The JCQ guidance, revised on 30 April 2025, does not tell students to stay away from AI. It tells them to leave a trail.

Three things, if a student uses an AI tool in work that counts. Reference it as you would a book, naming the tool and the date it produced the content, and the guidance gives its own example: “ChatGPT 3.5 (https://openai.com/blog/chatgpt/), 25/01/2025”. Keep the question you asked and the answer you got in a form you cannot edit later, so a screenshot will do. Then write a brief explanation of how you used it, handed in with the work.

Notice what that is not. Not a detection tool, not a permission slip. It is a requirement to show your working with the machine, so a teacher can read the AI output next to the finished piece and judge who did the thinking.

Why “show your working” beats “allowed or not allowed”

Rules about tools rot fast. Name ChatGPT in a policy and you are out of date by the next school holiday, and a blanket ban only moves the behaviour somewhere you cannot see. A rule about process holds up, because it asks about the child rather than the software.

Ofqual’s advice to teachers is pleasingly low tech. “You know your students. If something doesn’t sound like them, question it.” Its lesson plan hands students the test in a form a fifteen-year-old can answer honestly: does this work represent your knowledge, your skills, your thinking? On the Ofqual blog the same day, Amanda Swann, the regulator’s Executive Director of Delivery, put it plainly: “AI should not be used to generate coursework without proper disclosure. This is cheating, and the consequences, which can include losing a qualification entirely, are serious.” Ofqual also reports that 55% of teachers plan to ask students to explain their work if they have concerns.

That is the whole system. Explain your work. Everything else is paperwork around it.

The stakes, without the scare

Ofqual’s malpractice statistics for summer 2025, published on 11 December 2025, record 5,025 proven cases of student malpractice, with 4,735 students penalised, or 0.3% of the 1,376,480 candidates who received results. Worth saying clearly: those statistics do not break out AI as a category, so nobody can honestly point at them and claim an AI wave.

The cost when it does happen is not small. Ofqual’s briefing pack tells school leaders that 2025 penalties included 1,125 cases in which a student lost a whole GCSE or A level, and nearly 2,000 in which marks were deducted. The line it suggests schools use with parents is unsentimental: “Qualifications are important for your child’s future. Don’t let them miss out on the grades they deserve.”

The exposure, meanwhile, is ordinary rather than exotic. Internet Matters, surveying 1,000 UK children aged 9 to 17 and reporting in July 2025, found 64% had used an AI chatbot, and 42% of those who use one use it for schoolwork. Not a fringe of rule-breakers. Most of a year group, doing something nobody taught them to do properly.

Borrow the rule for your kitchen table

The British rule scales down nicely. You do not need an exam board to run it. Four habits, a minute each.

• Say it out loud. House rule: if AI touched the work, it gets mentioned. Which tool, what for. An entry, not a confession, so there is nothing to hide.
• Keep the screenshot. Ask for the prompt and the reply, saved. A child re-reading their own prompt a week later learns more about prompting than any lesson delivers.
• Ask the one-line question. “What did you do with it?” is the entire audit. A child who can answer has used the tool. A child who cannot has been used by it.
• Read the two versions side by side. Their draft, then the AI’s. Where their voice disappears is where the thinking got handed over. Point at it gently, then hand the paragraph back.

The nearby skill, if they cannot yet tell a tool that helped from a tool that took over, is how to teach your child to fact-check an AI. Build that one first.

The read from Singapore

Singapore reached a similar place by another road. In a parliamentary reply on 6 May 2026, MOE said national exams “continue to be proctored, and use of AI is prohibited”, and that where AI is permitted, “such as for coursework, Teacher Supervisors monitor students’ work such that AI use meets the objectives of the assessment”. Lower primary is kept clear of it entirely, and P4 to P6 use only education-designed tools under teacher supervision.

Supervision here does the work that declaration does in England. The gap is what happens at home, at nine at night, with a laptop and a deadline and no Teacher Supervisor standing by. That part is yours. Europe is pushing at the underlying skill: the European Commission and the OECD launched a joint AI Literacy Framework for primary and secondary schools in June 2026, organised around engaging with AI, creating with it, managing it and shaping it.

One disclosure: Mentus AI builds AI mentors for children, so we have an interest in coaching tools looking better than answering ones.

Start tonight with the smallest version. Not “did you use AI”, which invites a no. Try “show me what you asked it”. A child who can walk you through the question they typed and what they did with the answer has already done the harder half of the work, and has just proved it without either of you having to argue.